One way of proceeding is to post-doc in fields critically related. I
suggest hanging around with Karl Pribram (BRAINS Center, Radford, VA) --
who has written extensively on the neurophysiology and has even worked
with neural network types (like me).
We don't even have to worry about computational complexity of
dendro-dendritic and axodendritic networks to get nervous: talk about
chemical communication means and electrochemical field effects. These
are, as observed, very hard problems.
But the only effective way to pose questions in such a complex field is by
proposing models of systems -- they, at least, are provably wrong (which
all will be for the next hundred years...).
It may be worth suggesting, though, that higher-level models, based on
observed phenomema, may offer some sensible way to ask questions about
behavior (knowing that they, too, are doomed to be provably wrong). See
Leven & Levine, COG SCI (1996), Issue 2 (in press).